OHI British Columbia | OHI Science | Citation policy
OHI British Columbia | OHI Science | Citation policy
knitr::opts_chunk$set(fig.width = 6, fig.height = 4, fig.path = 'Figs/',
echo = TRUE, message = FALSE, warning = FALSE)
library(ohicore) ### devtools::install_github('ohi-science/ohicore')
source('~/github/ohibc/src/R/common.R')
dir_ohibc <- '~/github/ohibc'
dir_calc <- file.path(dir_ohibc, 'calc_ohibc')
dir_master <- file.path(dir_calc, 'master')
source(file.path(dir_calc, 'calc_scores_fxns.R'))
### provenance tracking
# library(provRmd); prov_setup()For each goal, regress the future status (default is +5 years) against current status, trend, pressures, and resilience; this would give an idea of coefficients in the OHI LFS model; these may differ for each goal. These do not apply to supragoals (since no pressures/resilience applied at this level), only standalone and subgoals.
\[ Status_{future} = Status_{current} [1 + \beta T + (1 - \beta) (R - P)]\] Rearrange to \[ \frac{Status_{future}}{Status_{current}} = 1 + \beta T_t + (1 - \beta)R_t - (1 - \beta)P_t\] In the original OHI publication, the coefficient \(\beta\) was set to 0.67, assuming that recent trend is a better predictor than pressures and resilience. To ensure an intuitive comparison, adjust the values in scores.csv so that \(T_t \in [-1, 1]\), and \(R_t, P_t \in [0, 1]\).
Framing as regression formula, allowing resilience and pressure coefficients to be estimated independently:
\[ X = \frac{Status_{t+lag}}{Status_{t}} = \beta_0 + \beta_1 T_t + \beta_2 R_t + \beta_3 P_t\] The standard OHI LFS model should result in parameter estimates: \(\beta_0 = 1, \beta_1 = .67, \beta_2 = .33, \beta_3 = -.33\).
Framing as regression where resilience and pressure are estimated together:
\[ X = \frac{Status_{t+lag}}{Status_{t}} = \beta_0 + \beta_1 T_t + \beta_2 (R_t - P_t)\] The standard OHI LFS model should result in parameter estimates: \(\beta_0 = 1, \beta_1 = .67, \beta_2 = .33\).
Framing as a trend-only model:
\[ X = \frac{Status_{t+lag}}{Status_{t}} = \beta_0 + \beta_1 T_t\]
Assuming region scores should follow roughly similar patterns, we can include all regions for each year in the regression. Clustering by region would be an alternative.
We will loop over the data by goal, but also by variable lag time in case pressures and resilience signals better predict a future state sooner or later than five years… in this case, then, trend should also be adjusted to the new lag since it is generally calculated to predict five years out.
### Set up basic data frame - make sure prs and res are 0-1; status
### range not important here since we'll take a ratio inside a loop.
scores_rgn <- read_csv(file.path(dir_calc, 'scores_all.csv')) %>%
filter(region_id != 0) %>%
spread(dimension, score) %>%
select(goal, region_id, year,
pressures, resilience,
status, trend) %>%
group_by(goal, region_id) %>%
arrange(year) %>%
mutate(pressures = pressures / 100,
resilience = resilience / 100,
res_minus_prs = resilience - pressures) %>%
filter(!is.na(pressures) & !is.na(resilience) &!is.na(trend))
lead_yrs <- 1:6
goals <- c('AO',
'FIS', 'SAL', # 'MAR', ### commented out - too few years
'CPP', 'CSS',
'CW',
'HAB', 'SPP',
'ICO', 'LSP',
'LEF', 'LEO',
'TR')
lfs1_all <- data.frame()
lfs2_all <- data.frame()
tr_only_all <- data.frame()
for(lead_yr in lead_yrs) { # lead_yr <- 5
for(goalname in goals) { # goalname <- goals[11]
# cat(goalname, ': lag =', lead_yr, '\n')
tmp_df <- scores_rgn %>%
filter(goal == goalname) %>%
mutate(obs_future_status = lead(status, lead_yr),
lfs_ratio = obs_future_status/status,
trend_adj = trend * lead_yr / 5) %>%
filter(!is.na(lfs_ratio))
# tmp_plot <- ggplot(tmp_df, aes(x = status, y = obs_future_status, color = year)) +
# ggtheme_plot() +
# geom_abline(slope = 1, intercept = 0, color = 'red') +
# geom_point(alpha = .5) +
# labs(x = 'current status', y = 'observed future status', title = goalname)
#
# print(tmp_plot)
lfs1_lm <- lm(lfs_ratio ~ trend_adj + pressures + resilience, data = tmp_df) %>%
broom::tidy() %>%
mutate(goal = goalname,
n_obs = nrow(tmp_df),
years = length(tmp_df$year %>% unique()),
lfs_lead = lead_yr)
lfs2_lm <- lm(lfs_ratio ~ trend_adj + res_minus_prs, data = tmp_df) %>%
broom::tidy() %>%
mutate(goal = goalname,
n_obs = nrow(tmp_df),
years = length(tmp_df$year %>% unique()),
lfs_lead = lead_yr)
tr_only_lm <- lm(lfs_ratio ~ trend_adj, data = tmp_df) %>%
broom::tidy() %>%
mutate(goal = goalname,
n_obs = nrow(tmp_df),
years = length(tmp_df$year %>% unique()),
lfs_lead = lead_yr)
lfs1_all <- bind_rows(lfs1_all, lfs1_lm)
lfs2_all <- bind_rows(lfs2_all, lfs2_lm)
tr_only_all <- bind_rows(tr_only_all, tr_only_lm)
}
}
lfs1_all <- lfs1_all %>%
mutate(estimate = round(estimate, 4),
std.error = round(std.error, 4),
statistic = round(statistic, 4),
p.value = round(p.value, 4))
lfs2_all <- lfs2_all %>%
mutate(estimate = round(estimate, 4),
std.error = round(std.error, 4),
statistic = round(statistic, 4),
p.value = round(p.value, 4))
tr_only_all <- tr_only_all %>%
mutate(estimate = round(estimate, 4),
std.error = round(std.error, 4),
statistic = round(statistic, 4),
p.value = round(p.value, 4))
write_csv(lfs1_all, 'int/lfs1_all.csv')
write_csv(lfs2_all, 'int/lfs2_all.csv')
write_csv(tr_only_all, 'int/trend_only_all.csv')lfs1_all <- read_csv('int/lfs1_all.csv') %>%
mutate(p_sig = case_when(p.value < .001 ~ 'p < .001',
p.value < .01 ~ 'p < .01',
p.value < .05 ~ 'p < .05',
TRUE ~ 'p >.05'),
p_sig = factor(p_sig, levels = c('p >.05', 'p < .05', 'p < .01', 'p < .001')))
for(goalname in goals) {
lfs1_goal <- lfs1_all %>%
filter(goal == goalname)
term_plot <- ggplot(lfs1_goal, aes(x = lfs_lead, y = estimate,
color = p_sig)) +
ggtheme_plot() +
geom_hline(yintercept = 0, color = 'red4', alpha = .5) +
geom_point(alpha = .8, size = 3) +
scale_x_continuous(breaks = 1:6, limits = c(.5, 7)) +
scale_color_manual(values = c('p >.05' = 'grey80', 'p < .05' = 'orange',
'p < .01' = 'yellowgreen', 'p < .001' = 'green4')) +
geom_text(aes(label = p_sig), nudge_x = .25, hjust = 0, color = 'grey20', size = 2) +
facet_wrap( ~ term, scales = 'free') +
labs(x = 'Future lag years',
y = 'Parameter value',
title = paste0(goalname, ': trend, pressures, resilience model'),
size = 'Significance')
print(term_plot)
}
lfs2_all <- read_csv('int/lfs2_all.csv') %>%
mutate(p_sig = case_when(p.value < .001 ~ 'p < .001',
p.value < .01 ~ 'p < .01',
p.value < .05 ~ 'p < .05',
TRUE ~ 'p >.05'),
p_sig = factor(p_sig, levels = c('p >.05', 'p < .05', 'p < .01', 'p < .001')))
for(goalname in goals) {
lfs2_goal <- lfs2_all %>%
filter(goal == goalname)
term_plot <- ggplot(lfs2_goal, aes(x = lfs_lead, y = estimate,
color = p_sig)) +
ggtheme_plot() +
geom_hline(yintercept = 0, color = 'red4', alpha = .5) +
geom_point(alpha = .8, size = 3) +
scale_x_continuous(breaks = 1:6, limits = c(.5, 7)) +
scale_color_manual(values = c('p >.05' = 'grey80', 'p < .05' = 'orange',
'p < .01' = 'yellowgreen', 'p < .001' = 'green4')) +
geom_text(aes(label = p_sig), nudge_x = .25, hjust = 0, color = 'grey20', size = 2) +
facet_wrap( ~ term, scales = 'free') +
labs(x = 'Future lag years',
y = 'Parameter value',
title = paste0(goalname, ': trend, pressures, resilience model'),
size = 'Significance')
print(term_plot)
}
tr_only_all <- read_csv('int/trend_only_all.csv') %>%
mutate(p_sig = case_when(p.value < .001 ~ 'p < .001',
p.value < .01 ~ 'p < .01',
p.value < .05 ~ 'p < .05',
TRUE ~ 'p >.05'),
p_sig = factor(p_sig, levels = c('p >.05', 'p < .05', 'p < .01', 'p < .001')))
for(goalname in goals) {
tr_only_goal <- tr_only_all %>%
filter(goal == goalname)
term_plot <- ggplot(tr_only_goal, aes(x = lfs_lead, y = estimate, color = p_sig)) +
ggtheme_plot() +
geom_hline(yintercept = 0, color = 'red4', alpha = .5) +
geom_point(alpha = .8, size = 3) +
scale_x_continuous(breaks = 1:6, limits = c(.5, 7)) +
scale_color_manual(values = c('p >.05' = 'grey80', 'p < .05' = 'orange',
'p < .01' = 'yellowgreen', 'p < .001' = 'green4')) +
geom_text(aes(label = p_sig), nudge_x = .25, hjust = 0, color = 'grey20', size = 2) +
facet_wrap( ~ term, scales = 'free') +
labs(x = 'Future lag years',
y = 'Parameter value',
title = paste0(goalname, ': trend only model'),
size = 'Significance')
print(term_plot)
}
